Argonne National Laboratory

The New Paradigm of Petascale Computing

(Page 2 of 3)

“There’s a multistep process to get codes ready for the BlueGene/P,” says Stevens.  “If they are already running on the BlueGene/L then they are in pretty good shape.” He says there are currently about 100 applications that run on BlueGene/L, ranging from astrophysics to chemistry to economics. But some application scientists will need to modify their software to get ready for the BlueGene/P.

BlueGene/P Short List

Stevens has a short list of applications that are likely to be the first to access the BlueGene/P system.  One such application is the earthquake simulations led by Arthur Rodgers, Anders Petersson and David McCallen of Lawrence Livermore National Laboratory (LLNL).

As the U.S. searches for new sources of energy, there is renewed interest in building new nuclear power plants. But safety concerns about earthquake hazards, particularly in the western U.S., require detailed studies of potential nuclear plant sites.

Ground motions computed with Wave Propagation Program code along two points above a hypothetical magnitude 7.0 earthquake Click image for larger version and more information

“Ideally, one would like to be able to model how an earthquake fault might rupture, how those seismic waves may propagate through the earth, and how they may arrive at your site,” says McCallen, LLNL division leader in nonproliferation, homeland and international security.  Historically, such predictions have been made based on past earthquakes and then extrapolated to predict future earthquake damage.  But these methods have limitations.  “You can make better predictions about the ground movements at a given site from physics and what we call ‘first principles.’” The problem is that combining data from subsurface geology and earthquake fault models has been prohibitive because there simply has not been enough computational power available to do the simulations.

“This is an application that demands as much computer crunching and computer power as you can muster up,” he says.  “It’s a heavily three-dimensional computationally intensive problem.  Even today, our biggest computers will be taxed by this application.  That’s a good example of why there is a motivation to use the emerging leadership computing facilities at Argonne.” With the additional computing power, McCallen would like to construct fault models of an area of 50 to 100 kilometers across by 80 kilometers deep and then combine that data with subsurface geologic models to simulate what the ground motion might be at a given site for different earthquake scenarios.

To do that, the group uses Wave Propagation Program (WPP) developed through laboratory-directed research at Lawrence Livermore, a program that is expected to scale up to petascale relatively easily, since it has already proven it runs efficiently on a massively parallel computing architecture.

An Argonne group, led by Bob Hill, is working on next-generation nuclear power plant design, so there is a natural link between the groups, says McCallen.

« Previous       1   |   2   |   3   |   References   |   Print       Next »